MITACS Globalink-funded AI/ML Engineer and Researcher specializing in agentic LLM systems, retrieval-augmented generation, and multimodal computer vision. Shipped 10+ production AI systems and a first-author manuscript on hierarchical preference graphs and federated LoRA adaptation for personalized LLM agents; designed a novel Cross-View Attention architecture achieving 83% AUC across 14 respiratory pathologies. Combines end-to-end ownership — from model architecture to full-stack deployment — with a track record of building explainable, privacy-preserving AI for healthcare and clinical decision support.
Get to know me better
MITACS Globalink-funded AI/ML Engineer and Researcher specializing in agentic LLM systems, retrieval-augmented generation, and multimodal computer vision. Shipped 10+ production AI systems and a first-author manuscript on hierarchical preference graphs and federated LoRA adaptation for personalized LLM agents; designed a novel Cross-View Attention architecture achieving 83% AUC across 14 respiratory pathologies. Combines end-to-end ownership — from model architecture to full-stack deployment — with a track record of building explainable, privacy-preserving AI for healthcare and clinical decision support.
With a background spanning freelance engineering and corporate innovation, I’ve delivered range of full-stack AI projects, spanning from smart database querying systems to automated medical assessment tools. I’ve led award-winning team in computer vision and am currently pushing the boundaries of Federated Learning—a way for AI to learn from sensitive data without ever compromising user privacy.
Bachelor of Science in Computer Science University of Engineering and Technology (UET), Lahore
Download my latest resume
View my qualifications, experience, and skills
My professional journey
My academic background
Thesis: AiroDx — Federated Context-Aware Explainable AI Framework for Multimodal Differential Diagnosis of Respiratory Diseases
Things I've built
Developed a sophisticated diagnostic system that synchronizes spatial features from multiple X-ray perspectives using custom attention mechanisms and self-supervised ViT backbones. The pipeline automates the complex mapping of granular PadChest labels to target pathologies while managing massive datasets through efficient filename indexing and multi-worker loading. Optimized model convergence using a two-stage training strategy—transitioning from frozen feature extraction to full-model fine-tuning—to maximize per-class AUC and F1 scores.
Developed a full-stack AI SaaS platform that interprets medical lab reports using a 5-stage LLM pipeline (OCR → retrieval → analysis → trend detection → infographics), simplifying complex clinical data for patients and non-experts. Engineered a scalable and secure FastAPI backend with JWT authentication, async MongoDB, Pinecone vector search, rate limiting, and structured logging to deliver robust AI inference services. Deployed the solution with a Next.js frontend on Vercel and FastAPI backend on Hugging Face Spaces, providing a cloud-hosted service with file uploads, subscription management, and real-time report analysis.
Developed secure AI backend systems for healthcare predictions using a Zero Trust architecture to protect model integrity and data access. Implemented cryptographic verification and security mechanisms that significantly reduced simulated attack success rates.
Designed a preference-aware LLM agent that models user preferences using hierarchical graphs and dynamic LoRA adapter routing for personalized responses. The system incorporates federated continual learning to improve adaptation while maintaining privacy across distributed environments.
Jafs Gjerdrum is a comprehensive digital solution designed to modernize restaurant operations and enhance customer experience. This full-stack web application seamlessly integrates online ordering, real-time order management, and administrative controls into a single, user-friendly platform.
Implemented a hybrid recommendation system combining transformer-based text embeddings with neural collaborative filtering for personalized movie suggestions. The model improved recommendation quality and addressed cold-start problems through integrated content and interaction signals.
Developed a fully automated pipeline that generates motivational YouTube Shorts using LLM-based content generation, text-to-speech voiceovers, and stock footage from Pexels. The system programmatically blends video, captions, and audio using MoviePy and supports automated uploading or scheduling via the YouTube Data API.
This project aims to determine whether a pair of questions from Quora are duplicates. Given a dataset of question pairs, various machine learning models were trained and evaluated to achieve the highest accuracy in predicting duplicate questions. The final model chosen was a Random Forest classifier due to its superior performance compared to other models.
A Node.js CLI + GUI tool that analyzes Upwork job listings against your portfolio to evaluate alignment, budget fit, and hiring chances. It also leverages AI to generate highly personalized, job-specific proposals, streamlining your freelancing workflow.
Research & writings
Professional credentials
Deeplearning.ai (Coursera)
Deeplearning.ai (Coursera)
Milestones and recognitions
Selected competitively among top-ranked international students for a 12-week funded research internship at the University of Winnipeg, Canada. Project title: Integrating an LLM into a next-generation smartwatch health application for in-situ data exploration.
Ranked #1 out of approximately 55 industry and research-grade projects with AiroDx. Project evaluated by the internals (professors) and externals (senior industry experts).
Top team out of 10+ competing teams, solving real world problems. Details about the project is in experience section
Coordinated protocol, graphics, and event management as part of the leadership team for Pakistan’s premier free hackathon, serving 300+ participants.
What people say about me
“Working with Omar was an absolute pleasure! He’s professional, skilled, and incredibly responsive throughout the entire project. What impressed me most was his attention to detail and ability to truly understand my vision ,he not only delivered exactly what I wanted but even improved it with his own creative touch. Communication was smooth, deadlines were met without any issues, and the quality of the final result exceeded my expectations. I highly recommend Omar to anyone looking for a reliable and talented developer. I will definitely be working with him again!”
Visionary Architect · ViewAI
“I have worked with Omar on multiple projects, and every time he has been outstanding. He is extremely professional and consistently delivers high-quality work. Omar understands the requirements very quickly—sometimes even before you fully explain them—and always exceeds expectations. Working with him has been a great experience, and I highly recommend him to anyone looking for reliable and exceptional work.”
King Fahad University
“The development of Jafsgjerdrum exceeded our expectations in every aspect. The platform delivers a seamless user experience while providing powerful administrative tools that have streamlined our operations significantly. Its reliability, performance, and thoughtful design make it an invaluable asset to our business. We highly recommend this service to anyone seeking a high-quality digital solution.”
CEO · Jafs Gjerdrum
“Omar brings a rare combination of research mindset, technical expertise, and engineering execution. His work in AI research, including LLM-based systems and multimodal deep learning, demonstrates his ability to solve complex problems and build impactful solutions with a strong focus on innovation.”
Associate Professor · UET Lahore
Let's connect and collaborate
